LANGPHY SERVICES - BILLING AUDIT REPORT

Your audit uses $0.50/GB/month for memory, but our actual RAM pricing is $10/GB/month (billed per minute). At ~4.5 GB of memory across your services, that alone accounts for roughly $45/month. Your most recent invoice (May 6) confirms this: $35.59 in memory charges, $3.95 in CPU, $1.58 in agent usage, $0.22 in network egress, and $0.11 in disk, plus the $5 Hobby subscription. Your current billing period usage so far is $15.07, with Kafka ($3.39), NLP ($2.93), speech-worker ($1.61), and speech-api ($1.49) as the top contributors, almost entirely from memory. You can review the per-service breakdown at your workspace usage page, and full pricing details are at docs.railway.com/pricing.

I saw an estimated cost of $30.36 yesterday morning. Then, in the afternoon, it suddenly went to $51+ without any real spike in usage. How did that happen? My 18 services RAM is limited to 0.5GB, and 4 services have 2 GB each. I can't drag them down because I have the 'Hobby' package. My CPU usage for 19 services is under 0.1 or around. But I have to keep them at 0.5. Now, please do a thorough audit, and please tell me what to do from my end to keep the cost stable!

Your current billing period (May 6 - Jun 6) has accumulated $15.07 in 9 days, and the dashboard projects that rate to the full month, reaching ~$51. The earlier $30 estimate reflected fewer days being extrapolated, and the May 11 memory spike during migration temporarily inflated the projection further. Memory is $12.63 of that $15.07: Kafka ($3.32), NLP ($2.93), speech-worker ($1.60), speech-api ($1.49), and langphy-web in another project ($1.30) are the top contributors. Your 19 smaller services add ~$4.33 combined, mostly memory, because RAM is $10/GB/month (not $0.50 as in your audit). To control costs, enable Serverless on services that don't need to be always-on (stops them when idle, eliminating memory charges during downtime), set replica limits to cap per-service memory, or configure a hard usage limit to enforce a spending ceiling. Your $5 Hobby subscription covers the first $5 of usage, so your final bill equals whichever is greater: $5 or total resource usage.

I didn't find the option to partially stop a service if it doesn't do anything. caddy sleeps automatically, but how to do that with others? can you guide me?

RAM is $10/GB/month, not $0.50/GB/month. At ~4.5 GB total, that's ~$45 in memory alone, which matches your projected bill. No discrepancy.

To cut costs:

• Enable Serverless per service (Settings > Deploy > Serverless)
• Set per-service replica limits to cap memory
• Set a hard usage limit to enforce a spending ceiling

Pricing: https://docs.railway.com/pricing

Thank you! One more question. If I want to move to the Pro plan ($20), can I set RAM and CPU limits per service to whatever I want? Now I can only set the CPU and RAM to 0.5vCPU and 0.5GB, respectively, not 0.1vCPU and 100MB RAM per service. And how will the bill be counted then?

On Pro ($20/mo), the $20 subscription covers your first $20 of usage, and you only pay more if total usage exceeds that. Resource pricing is identical to Hobby ($10/GB/month for RAM, $20/vCPU/month for CPU, billed per minute for actual consumption). Pro raises the per-service ceiling to 32 vCPU and 32 GB, but the minimum replica limit you can set is plan-controlled and not lower on Pro than on Hobby. Replica limits cap the maximum a service can consume, not what you're billed for - you're billed for actual usage only. The most effective way to reduce your bill is enabling Serverless on services that sit idle, since that stops them entirely when inactive and eliminates their memory charges. The upgrade is pro-rated, so unused Hobby credit applies automatically.